Sei sulla pagina 1di 8

Annals of Internal Medicine ORIGINAL RESEARCH

Reconciling the Effects of Screening on Prostate Cancer Mortality in the


ERSPC and PLCO Trials
Alex Tsodikov, PhD; Roman Gulati, MS; Eveline A.M. Heijnsdijk, PhD; Paul F. Pinsky, PhD; Sue M. Moss, PhD; Sheng Qiu, MS;
Tiago M. de Carvalho, MS; Jonas Hugosson, MD; Christine D. Berg, MD; Anssi Auvinen, MD; Gerald L. Andriole, MD;
Monique J. Roobol, PhD; E. David Crawford, MD; Vera Nelen, MD; Maciej Kwiatkowski, MD; Marco Zappa, PhD;
Marcos Lujan, MD; Arnauld Villers, MD; Eric J. Feuer, PhD; Harry J. de Koning, MD; Angela B. Mariotto, PhD; and
Ruth Etzioni, PhD

Background: The ERSPC (European Randomized Study of Results: Estimated MLTs were similar in the ERSPC and PLCO
Screening for Prostate Cancer) found that screening reduced intervention groups but were longer in the PLCO control group
prostate cancer mortality, but the PLCO (Prostate, Lung, Colorec- than the ERSPC control group. Extended analyses found no evi-
tal, and Ovarian Cancer Screening Trial) found no reduction. dence that effects of screening differed between trials (P = 0.37
to 0.47 [range across MLT estimation approaches]) but strong
Objective: To evaluate whether effects of screening on prostate evidence that benet increased with MLT (P = 0.0027 to 0.0032).
cancer mortality relative to no screening differed between the Screening was estimated to confer a 7% to 9% reduction in the
ERSPC and PLCO. risk for prostate cancer death per year of MLT. This translated
Design: Cox regression of prostate cancer death in each trial into estimates of 25% to 31% and 27% to 32% lower risk for
group, adjusted for age and trial. Extended analyses accounted prostate cancer death with screening as performed in the ERSPC
for increased incidence due to screening and diagnostic and PLCO intervention groups, respectively, compared with no
work-up in each group via mean lead times (MLTs), which were screening.
estimated empirically and using analytic or microsimulation Limitation: The MLT is a simple metric of screening and diag-
models. nostic work-up.
Setting: Randomized controlled trials in Europe and the United Conclusion: After differences in implementation and settings
States. are accounted for, the ERSPC and PLCO provide compatible ev-
Participants: Men aged 55 to 69 (ERSPC) or 55 to 74 (PLCO) idence that screening reduces prostate cancer mortality.
years at randomization. Primary Funding Source: National Cancer Institute.
Intervention: Prostate cancer screening.

Measurements: Prostate cancer incidence and survival from Ann Intern Med. doi:10.7326/M16-2586 Annals.org
randomization; prostate cancer incidence in the United States For author afliations, see end of text.
before screening began. This article was published at Annals.org on 5 September 2017.

M ore than 2 decades after prostate-specic antigen


(PSA) screening for prostate cancer entered clini-
cal practice, in 2012 the U.S. Preventive Services Task
3.0 g/L in most ERSPC centers and rounds), and
stopped regular screening after 6 rounds. Prostate can-
cer incidence was higher in the United States than in
Force (USPSTF) determined that there was very low Europe before the trials started, reecting different
probability of preventing a death from prostate cancer populations and clinical diagnosis patterns. The U.S.
in the long term and recommended against routine practice setting also contributed to more frequent
use of the test (1). Since then, rates of PSA screening screening in the PLCO control group and less frequent
and prostate cancer incidence have decreased signi- biopsy than in the ERSPC. Consequently, the PLCO
cantly in the United States (2, 3). compared the effects of an organized screening pro-
The USPSTF recommendation relied heavily on re- gram versus opportunistic screening rather than
sults from the ERSPC (European Randomized Study of screening versus no screening (8 10). Nonetheless, the
Screening for Prostate Cancer) (ISRCTN registry num- PLCO results have been viewed as more relevant to the
ber: ISRCTN49127736) and the PLCO (Prostate, Lung, U.S. setting (11).
Colorectal, and Ovarian Cancer Screening Trial) The objectives of this study were to formally test
(ClinicalTrials.gov: NCT00002540). However, the trials whether the effects of screening on prostate cancer
had apparently conicting results, with the ERSPC re-
porting a 21% reduction in prostate cancer mortality
(4 6) and the PLCO nding no difference in mortality See also:
between the trial groups (79). Interpretation of the trial Editorial comment . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
results is complicated by differences in their implemen-
Summary for Patients . . . . . . . . . . . . . . . . . . . . . . . . . 2
tation (including design and adherence) and practice
settings. The PLCO used a shorter screening interval Web-Only
(annual vs. every 2 to 4 years in the ERSPC), had a
Supplement
higher PSA threshold for biopsy referral (4.0 g/L vs.
Annals.org Annals of Internal Medicine 1

Downloaded From: https://annals.org/ by a Univ of California San Francisco User on 09/04/2017


ORIGINAL RESEARCH Effects of Screening on Prostate Cancer Mortality in ERSPC and PLCO

mortality differed between the ERSPC and PLCO after Of note, our estimates of the MLTs differ from other
differences in implementation and practice settings estimates in the literature that can be interpreted as the
were accounted for and to estimate the effects of average time by which screening advances diagnosis
screening in both trials relative to no screening. among cases that would have been clinically diag-
nosed (15). Our MLTs are designed as proxies for the
intensity of screening and diagnosis, with higher values
reecting higher attendance rates at screening exami-
METHODS nations, more frequent screening examinations, less
Overview conservative criteria for biopsy referral, and/or more
Our study used individual records from both trials frequent biopsy. Thus, accounting for variable MLTs
in a collaboration between trial investigators and the across trial groups captures in a single measure impor-
prostate cancer working group of the Cancer Interven- tant differences in the trial screening protocols, partici-
tion and Surveillance Modeling Network. In the inter- pant adherence to those protocols in the intervention
vention groups, these records included age and year of groups, and control group screening.
randomization, enrollment center, dates and results of We estimated the MLTs empirically, with no model
PSA tests and rectal examinations, whether biopsy was assumptions about cancer progression and diagnosis,
performed, date of cancer diagnosis, and date and and also using 3 models of cancer natural history and
cause of death. In the control groups, the records in- diagnosis. The empirical approach estimated the MLTs
cluded age and year of randomization, enrollment cen- by calculating the difference between survival curves
ter, date of cancer diagnosis, and date and cause of for observed time from randomization to diagnosis in
death. For consistency with prior publications, ERSPC each trial group relative to an assumed baseline level.
data included men aged 55 to 69 years at randomiza- The assumed baseline probability of diagnosis in the
tion (12), and PLCO data included men aged 55 to 74 absence of screening was derived using incidence
years at randomization (13). rates from the SEER (Surveillance, Epidemiology, and
We rst conducted a traditional statistical analysis End Results) program in 1986 just before PSA screen-
that combined data from both trials and compared haz- ing began in the United Stateswith adjustment to re-
ards of prostate cancer death in the intervention ect distributions of age at randomization in each trial.
groups versus the control groups, with adjustment for In addition, 1 analytic model (University of Michigan
participant age and trial setting. However, this analysis [UMICH]) and 2 simulation models (Fred Hutchinson
is questionable because of remaining differences in im- Cancer Research Center [FHCRC] and Erasmus Univer-
plementation between the trials. To overcome this lim- sity Medical Center MIcrosimulation SCreening ANaly-
itation, we also performed extended analyses that ac- sis [MISCAN]) estimated times from randomization to
counted for variable screening and diagnostic work-up diagnosis in the absence and presence of screening
(hereafter screening intensity) in each trial group, based on cancer progression and diagnosis rates,
which we operationalized using mean lead times which were estimated using individual-patient data on
(MLTs). The MLTs reect the magnitude of increased attendance, screening, and incidence. The tted mod-
prostate cancer incidence relative to a baseline level els then estimated MLTs as in the empirical approach,
expected in the absence of screening, thus capturing but using projected instead of observed incidence
differences in both design and adherence (see the next rates. Each MLT was then scaled by the corresponding
section). We estimated the MLTs both empirically and fraction of patients diagnosed within the 11-year
using analytic or microsimulation models; using multi- follow-up and was projected so that it could be inter-
ple approaches allowed us to assess the robustness of preted as an average interval among cancer cases de-
results to this uncertain quantity. tected in the relevant trial group. Further details are
provided in the Supplement (available at Annals.org).
Estimating MLTs
Statistical Analysis
The MLT is usually dened as the average time by
We used Cox regression to model survival from
which diagnosis is advanced by screening relative to
randomization to prostate cancer death, censoring per-
the date of diagnosis without screening. Under com-
sons who died of other causes or were alive at the last
plete follow-up, where all preclinical cases are eventu-
follow-up. We performed both a traditional statistical
ally diagnosed in the no-screening setting, the MLT cor-
analysis and extended analyses that incorporated the
responds to the difference in areas under 2 survival
measure of screening intensity captured by the esti-
curves (one in the absence of screening minus one in
mated MLTs. Both types of analysis included participant
the presence of screening) for time from randomization
age at randomization and a trial setting indicator (PLCO
to diagnosis. Under limited follow-up, we can dene a
vs. ERSPC), which allowed for a different baseline risk
restricted version of the MLT as an analogous differ-
for prostate cancer death in the absence of screening
ence in areas under survival curves up to a specied between trial settings.
time point (14). Restricting the analysis to the duration
of the trial recognizes that events after the trial period
cannot affect mortality during the trial. To make esti- Traditional Statistical Analysis
mates between trials comparable, follow-up was re- We rst conducted a traditional analysis to test
stricted to 11 years. whether the effect of screening differed between trials.
2 Annals of Internal Medicine Annals.org

Downloaded From: https://annals.org/ by a Univ of California San Francisco User on 09/04/2017


Effects of Screening on Prostate Cancer Mortality in ERSPC and PLCO ORIGINAL RESEARCH
Specically, we tested the effect of being randomly as- tervention and control group) to a continuous metric of
signed to the intervention group (relative to the control screening intensity, with resulting coefcient estimates
group) on the risk for prostate cancer death. The expo- interpreted relative to a no-screening setting (where
nential of the coefcient for the trial group indicator is the MLT equals zero). Second, we used combined data
the hazard ratio for prostate cancer death in the inter- from both trials in a single analysis, adding an indicator
vention group relative to the control group; in other for trial to capture differences between trials in baseline
words, it reects the effect of screening on prostate cancer-specic survival without screening and an inter-
cancer mortality in an intention-to-screen analysis. We action term to test whether screening efcacy (per year
tted this model with and without allowing separate of MLT) differed between trials.
effects of screening in each trial (that is, with and with-
Role of the Funding Source
out interaction between the trial group and the trial
This study was supported by the National Cancer
setting indicator), then used a likelihood ratio test to
Institute, which had no role in the design, conduct, or
evaluate evidence of differential effects of screening
analysis of the study or the decision to submit the man-
between trials.
uscript for publication.

Extended Statistical Analysis


Next, we replaced the trial group indicator with the RESULTS
corresponding MLT, which was estimated empirically Table 1 summarizes participant characteristics,
or using a model-based approach. The exponential of follow-up, and prostate cancer cases and deaths in the
the coefcient for the MLT represents the hazard ratio 2 trials under all available follow-up and restricted to
for prostate cancer death per additional year of MLT; in 11 years of follow-up. The data under all available
other words, it reects screening efcacy standardized follow-up differ modestly from published results (5, 8)
by screening intensity. As in the traditional analysis, we because of cleaning and updating. Nonetheless, the
tted this model with and without allowing separate cleaned and updated data restricted to 11 years of
effects of screening on prostate cancer mortality in follow-up yielded values similar to published prostate
each trial (that is, with and without interaction between cancer incidence rate ratios (PLCO: 1.12 vs. 1.12;
the MLT and the trial setting indicator), then used a ERSPC: 1.68 vs. 1.63) and mortality rate ratios (PLCO:
likelihood ratio test to evaluate evidence of differential 1.02 vs. 1.09; ERSPC: 0.79 vs. 0.79) and preserved the
effects of screening between trials. greater effects of screening on prostate cancer inci-
Our extended analyses are consistent with the anal- dence and mortality rates in the ERSPC relative to the
yses in the trial publications (4, 7), with 2 important dif- PLCO.
ferences. First, rather than relying on an intention-to- To compare screening intensity in the intervention
treat effect of screening determined by the assigned and control groups of the trials, Figure 1 illustrates
group in a single trial, we explicitly included a covariate MLTs estimated empirically or using a model-based ap-
(MLT) to capture the intensity of screening in each proach. All estimation approaches found similar order-
group. This represents a transition from thinking about ing and relative magnitudes of MLTs across groups.
screening as all or nothing (corresponding to an in- The ERSPC and PLCO intervention groups had similar

Table 1. Summary of Participant Characteristics, Follow-up, and Prostate Cancer Cases and Deaths in the ERSPC and PLCO,
Under All Available Follow-up and Restricted to 11 Years of Follow-up

Characteristic ERSPC PLCO

Control Screening Control Screening


Participants, n 88 921 72 473 38 343 38 340
Median age at randomization (range), y 59 (5569) 60 (5569) 62 (5574) 62 (5574)
All available follow-up
Median follow-up from randomization (range), y 11.0 (0.417.5) 11.1 (0.417.3) 12.5 (013.0) 12.5 (013.0)
Prostate cancer cases, n 5398 6967 4040 4430
Person-years of follow-up for incidence 933 854 740 775 403 955 400 008
Deaths, n 17 019 13 652 7149 6940
Other causes 16 557 13 353 7003 6788
Prostate cancer 462 299 146 152
Person-years of follow-up for mortality 990 678 827 148 426 720 427 824
Restricted to 11 y of follow-up
Median follow-up from randomization (range), y 11.0 (0.411.0) 11.0 (0.411.0) 11.0 (011.0) 11.0 (011.0)
Prostate cancer cases, n 4961 6586 3641 4038
Person-years of follow-up for incidence 868 834 686 766 368 844 365 129
Deaths, n 13 207 10 397 5880 5798
Other causes 12 822 10 150 5771 5687
Prostate cancer 385 247 109 111
Person-years of follow-up for mortality 890 581 725 997 387 027 387 861
ERSPC = European Randomized Study of Screening for Prostate Cancer; PLCO = Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial.

Annals.org Annals of Internal Medicine 3

Downloaded From: https://annals.org/ by a Univ of California San Francisco User on 09/04/2017


ORIGINAL RESEARCH Effects of Screening on Prostate Cancer Mortality in ERSPC and PLCO

Figure 1. Estimated MLTs in the intervention and control groups of the ERSPC and PLCO relative to a hypothetical
no-screening setting (where MLT equals zero).

Approach Group ERSPC PLCO

Empirical Intervention 4.0 4.0

Control 1.6 3.1

FHCRC Intervention 4.0 4.1

Control 0.9 3.0

MISCAN Intervention 3.5 4.6

Control 0.7 3.4

UMICH Intervention 3.8 4.0

Control 1.7 3.1

5 4 3 2 1 0 5 4 3 2 1 0

MLT, y MLT, y

Estimated MLTs are visualized as increasing to the left to suggest the extent to which prostate cancer diagnosis is advanced by more intensive
screening and diagnostic work-up. ERSPC = European Randomized Study of Screening for Prostate Cancer; FHCRC = Fred Hutchinson Cancer
Research Center; MISCAN = Erasmus University Medical Center MIcrosimulation SCreening ANalysis; MLT = mean lead time; PLCO = Prostate,
Lung, Colorectal, and Ovarian Cancer Screening Trial; UMICH = University of Michigan.

MLTs, but the PLCO control group had substantially vention groups, respectively, over 11 years of follow-up
longer MLTs than the ERSPC control group, which is relative to no screening.
consistent with more intensive screening (that is, greater Figure 2 shows prostate cancer survival from ran-
contamination) in the PLCO control group. domization in each trial group, obtained by Kaplan
Table 2 shows results of the traditional analysis. A Meier estimation and predicted under a common effect
likelihood ratio test associated with this analysis sug- of screening given MLTs estimated by the empirical ap-
gested marginal evidence of different effects of screen- proach. Predictions obtained using MLTs estimated by
ing on mortality between trials (P for interaction = the model-based approaches (not shown) are similar.
0.087). Under a common effect of screening, screening The predicted curves closely reproduce observed dif-
was estimated to reduce the risk for prostate cancer ferences in prostate cancer survival between the inter-
death by 16% (95% CI, 4% to 27%; P = 0.010) after we vention and control groups in both trials, showing that
accounted for different baseline risks for prostate can- screening intensity as captured by the MLT is highly
cer death in the PLCO setting relative to the ERSPC informative about between-group differences in risks
setting and participant age at randomization. This result for prostate cancer death in both trials.
essentially corresponds to a weighted average of the
effect in each trial, with the relative sizes of the trials as
weights. DISCUSSION
Table 2 also presents our extended analyses, which The USPSTF is currently updating its recommenda-
account for the MLT in each trial group, estimated em- tions about PSA screening and has previously used the
pirically or using a model-based approach. The analy- ERSPC and PLCO as its main sources of evidence about
ses are highly consistent and indicate no evidence of screening benet. Primary publications from these
different effects of screening on mortality between tri- high-quality randomized controlled trials are indispens-
als (P for interaction = 0.37 to 0.47 [range across esti- able for evaluating causal effects of screening for pros-
mation approaches]). Under a common effect of tate cancer. Yet, analyses like the one in this article that
screening, all approaches indicated strong evidence attempt to overcome limitations of traditional statistical
that a longer MLT was associated with a lower risk for analyses complement the empirical trial ndings by
prostate cancer death after differential baseline risks providing information about whether the evidence
for prostate cancer death between trial settings and from the trials is compatible and about the expected
participant age at randomization were accounted for reduction in prostate cancer mortality relative to no
(P = 0.0027 to 0.0032). These analyses estimated that screening.
screening conferred a 7% to 9% lower risk for prostate Rather than comparing the trial groups as if they
cancer death per year of MLT. Using the formula 1 represented screened and nonscreened populations,
(hazard ratio)MLT, this would translate into estimated this study estimated the intensity of screening in each
25% to 31% and 27% to 32% reductions in the ex- group relative to no screening. This allowed us to for-
pected risk for prostate cancer death in the setting of mally assess whether screening effects differed be-
screening as performed in the ERSPC and PLCO inter- tween the trials when we accounted for differential
4 Annals of Internal Medicine Annals.org

Downloaded From: https://annals.org/ by a Univ of California San Francisco User on 09/04/2017


Effects of Screening on Prostate Cancer Mortality in ERSPC and PLCO ORIGINAL RESEARCH
screening intensity between groups in each trial. By de- vival (16). By quantifying screening efcacy as a func-
coupling screening intensity from trial group labels and tion of intensity, we projected that screening decreased
investigating how benet depends on screening inten- the expected risk for prostate cancer death in both
sity, we concluded that differences between the ERSPC PLCO groups.
and PLCO results are largely attributable to differences We used several approaches to estimate screening
in screening intensities between groups within each intensity. The empirical approach reected catchall dif-
trial. After nding no evidence of different effects of ferences in the risk for prostate cancer diagnosis be-
screening on prostate cancer mortality between trials tween groups and calculated the MLT most consistent
given the screening intensities, we estimated a com- with incidence in each group relative to a common
mon effect of screening on mortality using pooled data baseline level. In contrast, the model-based ap-
on 19 226 prostate cancer cases. The pooled estimate proaches explicitly accounted for trial protocols and
showed a highly signicant benet of screening. This is practice setting details that were known or could be
the rst time that data from both trials have been har- quantied, such as age distributions, enrollment and
nessed to estimate screening benet. attendance patterns, and screening and biopsy fre-
This analysis may have had insufcient power to quencies within each ERSPC center. As expected, the
detect a signicant difference in screening efcacy be- estimates were shorter than in other studies (15) owing
tween trials. Thus, although there is no evidence of dif- to the different estimation approach and because we
ferent screening efcacies, we cannot unequivocally restricted our analyses to 11 years of follow-up. In gen-
conclude that they were identical. Nevertheless, our eral, results were highly consistent across estimation
combined analysis of both trials allowed the most pow- approaches and suggest robustness of our conclusions
erful examination of this question to date. to these ways of estimating screening intensity. Al-
Our analysis indicated that the baseline risk for though quantifying the screening intensity using the
prostate cancer death differed between trials. This MLT was simple and natural for our problem, other
could be due to different incidence, stage distributions, more complicated measures are possible. These in-
and treatment patterns in the trial populations in the clude various standardized measures derived from an
absence of screening. Lower-than-expected mortality excess hazard of cancer diagnosis in a screened popu-
(relative to survival in the prePSA screening era) was lation versus an unscreened population, using recent
observed in the PLCO, which may reect healthier par- methods (17) and earlier literature on relative survival
ticipants or an era with improved disease-specic sur- analysis.

Table 2. Results of Traditional and Extended Cox Regression Analyses of Death From Prostate Cancer and Estimated Mortality
Reductions in the ERSPC and PLCO Intervention Groups Relative to No Screening

Covariate Cox Regression Analysis Estimated Mortality Reduction Relative to No Screening

Hazard Ratio P Value ERSPC Intervention Group PLCO Intervention Group


(95% CI)
MLT, y Reduction MLT, y Reduction
(95% CI), % (95% CI), %
Traditional analysis
PLCO setting* 0.53 (0.450.62) <0.001
Participant age at randomization 1.13 (1.111.14) <0.001
Randomization to intervention group 0.84 (0.730.96) 0.0099 NA 16 (427) NA 16 (427)

Extended analyses
Empirical
PLCO setting* 0.57 (0.480.67) <0.001
Participant age at randomization 1.13 (1.111.14) <0.001
MLT 0.92 (0.870.97) 0.0027 3.96 29 (1143) 4.02 29 (1144)
FHCRC
PLCO setting* 0.58 (0.490.69) <0.001
Participant age at randomization 1.13 (1.111.14) <0.001
MLT 0.93 (0.880.97) 0.0029 4.00 27 (1040) 4.10 27 (1041)
MISCAN
PLCO setting* 0.63 (0.510.77) <0.001
Participant age at randomization 1.13 (1.111.14) <0.001
MLT 0.92 (0.870.97) 0.0032 3.49 25 (938) 4.62 32 (1247)
UMICH
PLCO setting* 0.57 (0.480.68) <0.001
Participant age at randomization 1.13 (1.111.14) <0.001
MLT 0.91 (0.850.97) 0.0029 3.83 31 (1245) 4.01 32 (1247)
ERSPC = European Randomized Study of Screening for Prostate Cancer; FHCRC = Fred Hutchinson Cancer Research Center; MISCAN = Erasmus
University Medical Center MIcrosimulation SCreening ANalysis; MLT = mean lead time; NA = not applicable; PLCO = Prostate, Lung, Colorectal,
and Ovarian Cancer Screening Trial; UMICH = University of Michigan.
* Relative to the ERSPC setting to account for differential baseline risk for prostate cancer death.
Continuous covariate.

Annals.org Annals of Internal Medicine 5

Downloaded From: https://annals.org/ by a Univ of California San Francisco User on 09/04/2017


ORIGINAL RESEARCH Effects of Screening on Prostate Cancer Mortality in ERSPC and PLCO

Figure 2. Prostate cancer survival from randomization in the ERSPC and PLCO, estimated by KaplanMeier or Cox regression
model using mean lead time estimated with the empirical approach.

ERSPC PLCO
100.0 100.0
Prostate Cancer Survival, %
99.8 99.8

99.6 99.6

99.4 99.4

99.2 99.2

99.0 99.0
0 2 4 6 8 10 12 0 2 4 6 8 10 12
Years Since Randomization Years Since Randomization

Model Intervention Control


KaplanMeier
Cox model

ERSPC = European Randomized Study of Screening for Prostate Cancer; PLCO = Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial.

The nding that effects of screening on mortality efcacy (relative to no screening) differed between the
seemed consistent between trials after differences in trials after we accounted for differences in implementa-
implementation and practice setting were accounted tion and setting. Our estimation of the common effect
for corroborates results from other analyses. For exam- of screening suggests that it can signicantly reduce
ple, a prior investigation of the PLCO found that control the risk for prostate cancer death. However, as for all
group screening substantially limited the power of that interventions, the benet of screening must be
trial to detect a clinically important reduction in pros- weighed against its potential harms for informed clini-
tate cancer mortality (18). However, that study did not cal and shared decision making.
formally evaluate whether effects of screening on pros-
tate cancer mortality differed between the ERSPC and From University of Michigan, Ann Arbor, Michigan; Fred
PLCO when implementation and setting details are Hutchinson Cancer Research Center, Seattle, Washington;
taken into account. Erasmus Medical Center, Rotterdam, the Netherlands; Na-
A limitation of this study is that we did not explicitly tional Cancer Institute, Bethesda, Maryland; Queen Mary Uni-
account for differences between trials in characteristics versity of London, London, United Kingdom; Sahlgrenska Uni-
of cancer cases (for example, clinical stage or Gleason versity Hospital, Goteborg, Sweden; Johns Hopkins Medicine,
score) or primary treatments. Any differences in these Baltimore, Maryland; University of Tampere, Tampere, Fin-
factors between trials are accounted for in the trial- land; Washington University School of Medicine, St. Louis,
Missouri; University of Colorado, Denver, Colorado; Provin-
specic baseline risks for prostate cancer death. Also,
ciaal Instituut voor Hygiene, Antwerp, Belgium; Kantonsspital
the model-based approaches to estimate MLTs assume
Aarau, Aarau, Switzerland; Institute for Cancer Prevention,
that cancer cases are progressive, although they allow
Florence, Italy; Universidad Complutense de Madrid, Parla,
heterogeneity of risk for progression across patients. It Madrid, Spain; and Universite de Lille, Lille, France.
is impossible to know whether some cancer cases
could remain indolent indenitely or regress spontane- Disclaimer: The content is solely the responsibility of the au-
ously and permanently. However, all estimation ap- thors and does not necessarily represent the ofcial views of
proaches closely matched incidence trends in each trial the National Cancer Institute.
group. We also assumed that incidence in the absence
of screening was constant across calendar years before Grant Support: By National Cancer Institute Award number
and after the trials began, which is a simplication. We U01CA157224.
considered only the MLT as a surrogate for screening
intensity. Other metrics could have associations with Disclosures: Dr. Moss reports grants from the Prostate Cancer
risk for prostate cancer death that differ from those we Research Foundation and the European Association of Urol-
found. Finally, the estimated mortality reduction in each ogy during the conduct of the study. Dr. Auvinen reports per-
trial group was based on the assumption that the risk sonal fees from EPID Research and MSD outside the submit-
for death from any cause is small during follow-up. ted work. Dr. Kwiatkowski reports personal fees from Myriad,
In conclusion, taken together, the data from the ER- Astellas, and Janssen outside the submitted work. Dr. Lujan
SPC and PLCO do not provide evidence that screening reports a grant from Fondo de Investigacion Sanitaria during
6 Annals of Internal Medicine Annals.org

Downloaded From: https://annals.org/ by a Univ of California San Francisco User on 09/04/2017


Effects of Screening on Prostate Cancer Mortality in ERSPC and PLCO ORIGINAL RESEARCH
the conduct of the study. Dr. Etzioni reports personal fees 6. Schroder FH, Hugosson J, Roobol MJ, Tammela TL, Zappa M,
from GRAIL outside the submitted work and ownership of Nelen V, et al; ERSPC Investigators. Screening and prostate
equity in Seno. Authors not named here have disclosed no cancer mortality: results of the European Randomised Study of
conicts of interest. Disclosures can also be viewed at www Screening for Prostate Cancer (ERSPC) at 13 years of follow-up. Lan-
cet. 2014;384:2027-35. [PMID: 25108889] doi:10.1016/S0140
.acponline.org/authors/icmje/ConictOfInterestForms.do?msNum
-6736(14)60525-0
=M16-2586. 7. Andriole GL, Crawford ED, Grubb RL 3rd, Buys SS, Chia D, Church
TR, et al; PLCO Project Team. Mortality results from a randomized
Reproducible Research Statement: Study protocol: Not avail- prostate-cancer screening trial. N Engl J Med. 2009;360:1310-9.
able. Statistical code: Source code or runs using the FHCRC [PMID: 19297565] doi:10.1056/NEJMoa0810696
8. Andriole GL, Crawford ED, Grubb RL 3rd, Buys SS, Chia D, Church
model are available from Mr. Gulati (e-mail, rgulati@fredhutch
TR, et al; PLCO Project Team. Prostate cancer screening in the ran-
.org), runs using the MISCAN model are available from Dr. domized Prostate, Lung, Colorectal, and Ovarian Cancer Screening
Heijnsdijk (e-mail, e.heijnsdijk@erasmusmc.nl), and source Trial: mortality results after 13 years of follow-up. J Natl Cancer Inst.
code or runs using the UMICH model are available from Dr. 2012;104:125-32. [PMID: 22228146] doi:10.1093/jnci/djr500
Tsodikov (e-mail, tsodikov@umich.edu). Data set: PLCO data 9. Pinsky PF, Prorok PC, Yu K, Kramer BS, Black A, Gohagan JK, et al.
are available from the National Cancer Institute Cancer Data Extended mortality results for prostate cancer screening in the PLCO
Access System (https://biometry.nci.nih.gov/cdas). ERSPC trial with median follow-up of 15 years. Cancer. 2017;123:592-9.
data may be available from Dr. Moss (e-mail, s.moss@qmul.ac [PMID: 27911486] doi:10.1002/cncr.30474
.uk). 10. Berg CD. The Prostate, Lung, Colorectal and Ovarian Cancer
Screening Trial: the prostate cancer screening results in context. Acta
Oncol. 2011;50 Suppl 1:12-7. [PMID: 21604935] doi:10.3109/0284
Requests for Single Reprints: Ruth Etzioni, PhD, Division of 186X.2010.531283
Public Health Sciences, Fred Hutchinson Cancer Research 11. Melnikow J, LeFevre M, Wilt TJ, Moyer VA. Counterpoint: ran-
Center, 1100 Fairview Avenue North, M2-B230, PO Box domized trials provide the strongest evidence for clinical guidelines:
19024, Seattle, WA 98109; e-mail, retzioni@fredhutch.org. the US Preventive Services Task Force and prostate cancer screen-
ing. Med Care. 2013;51:301-3. [PMID: 23481031] doi:10.1097/MLR
.0b013e31828a67d3
Current author addresses and author contributions are avail- 12. De Koning HJ, Hakulinen T, Moss SM, Adolfsson J, Smith PH,
able at Annals.org. Alexander FE; ERSPC. Monitoring the ERSPC trial. BJU Int. 2003;92
Suppl 2:112-4. [PMID: 14983967]
13. Prorok PC, Andriole GL, Bresalier RS, Buys SS, Chia D, Crawford
ED, et al; Prostate, Lung, Colorectal and Ovarian Cancer Screening
References Trial Project Team. Design of the Prostate, Lung, Colorectal and
1. Moyer VA; U.S. Preventive Services Task Force. Screening for Ovarian (PLCO) Cancer Screening Trial. Control Clin Trials. 2000;21:
prostate cancer: U.S. Preventive Services Task Force recommenda- 273S-309S. [PMID: 11189684]
tion statement. Ann Intern Med. 2012;157:120-34. [PMID: 14. Uno H, Claggett B, Tian L, Inoue E, Gallo P, Miyata T, et al. Mov-
22801674] doi:10.7326/0003-4819-157-2-201207170-00459 ing beyond the hazard ratio in quantifying the between-group differ-
2. Jemal A, Fedewa SA, Ma J, Siegel R, Lin CC, Brawley O, et al. ence in survival analysis. J Clin Oncol. 2014;32:2380-5. [PMID:
Prostate cancer incidence and PSA testing patterns in relation to 24982461] doi:10.1200/JCO.2014.55.2208
USPSTF screening recommendations. JAMA. 2015;314:2054-61. 15. Draisma G, Etzioni R, Tsodikov A, Mariotto A, Wever E, Gulati R,
[PMID: 26575061] doi:10.1001/jama.2015.14905 et al. Lead time and overdiagnosis in prostate-specic antigen
3. Jemal A, Ma J, Siegel R, Fedewa S, Brawley O, Ward EM. Prostate screening: importance of methods and context. J Natl Cancer Inst.
cancer incidence rates 2 years after the US Preventive Services Task 2009;101:374-83. [PMID: 19276453] doi:10.1093/jnci/djp001
Force recommendations against screening. JAMA Oncol. 2016;2: 16. Pinsky PF, Miller A, Kramer BS, Church T, Reding D, Prorok P,
1657-60. [PMID: 27541955] doi:10.1001/jamaoncol.2016.2667 et al. Evidence of a healthy volunteer effect in the Prostate, Lung,
4. Schroder FH, Hugosson J, Roobol MJ, Tammela TL, Ciatto S, Colorectal, and Ovarian Cancer Screening Trial. Am J Epidemiol.
Nelen V, et al; ERSPC Investigators. Screening and prostate-cancer 2007;165:874-81. [PMID: 17244633]
mortality in a randomized European study. N Engl J Med. 2009;360: 17. Sasieni P, Brentnall AR. On standardized relative survival. Biomet-
1320-8. [PMID: 19297566] doi:10.1056/NEJMoa0810084 rics. 2017;73:473-82. [PMID: 27554303] doi:10.1111/biom.12578
5. Schroder FH, Hugosson J, Roobol MJ, Tammela TL, Ciatto S, 18. Gulati R, Tsodikov A, Wever EM, Mariotto AB, Heijnsdijk EA,
Nelen V, et al; ERSPC Investigators. Prostate-cancer mortality at 11 Katcher J, et al. The impact of PLCO control arm contamination on
years of follow-up. N Engl J Med. 2012;366:981-90. [PMID: perceived PSA screening efcacy. Cancer Causes Control. 2012;23:
22417251] doi:10.1056/NEJMoa1113135 827-35. [PMID: 22488488] doi:10.1007/s10552-012-9951-8

Annals.org Annals of Internal Medicine 7

Downloaded From: https://annals.org/ by a Univ of California San Francisco User on 09/04/2017


Current Author Addresses: Dr. Tsodikov and Mr. Qiu: Depart- Author Contributions: Conception and design: A. Tsodikov, R.
ment of Biostatistics, University of Michigan, 1415 Washington Gulati, J. Hugosson, E.D. Crawford, A. Villers, E.J. Feuer, H.J.
Heights, Ann Arbor, MI 48109-2029. de Koning, R. Etzioni.
Mr. Gulati and Dr. Etzioni: Division of Public Health Sciences, Analysis and interpretation of the data: A. Tsodikov, R. Gulati,
Fred Hutchinson Cancer Research Center, 1100 Fairview Av- E.A.M. Heijnsdijk, P.F. Pinsky, S. Qiu, C.D. Berg, A. Auvinen,
enue North, Seattle, WA 98109-1024. G.L. Andriole, M. Zappa, A. Villers, H.J. de Koning, A.B.
Drs. Heijnsdijk and de Koning and Mr. de Carvalho: Depart- Mariotto, R. Etzioni.
ment of Public Health, Erasmus Medical Center, PO Box 2040, Drafting of the article: A. Tsodikov, R. Gulati, E.D. Crawford,
3000 CA Rotterdam, the Netherlands. H.J. de Koning, R. Etzioni.
Dr. Pinsky: Division of Cancer Prevention, National Cancer In- Critical revision of the article for important intellectual con-
stitute, 9609 Medical Center Drive, Bethesda, MD 20892. tent: A. Tsodikov, R. Gulati, E.A.M. Heijnsdijk, P.F. Pinsky, S.M.
Dr. Moss: Wolfson Institute, Queen Mary University of London, Moss, T.M. de Carvalho, C.D. Berg, A. Auvinen, G.L. Andriole,
Charterhouse Square, London EC1M 6BQ, United Kingdom. M.J. Roobol, M. Kwiatkowski, A. Villers, E.J. Feuer, H.J. de
Dr. Hugosson: Department of Urology, Sahlgrenska University Koning, R. Etzioni.
Hospital, Bl strket 5, 413 45 Goteborg, Sweden. Final approval of the article: A. Tsodikov, R. Gulati, E.A.M.
Dr. Berg: Department of Radiation Oncology and Molecular Heijnsdijk, P.F. Pinsky, S.M. Moss, S. Qiu, T.M. de Carvalho, J.
Radiation Sciences, Johns Hopkins Medicine, 401 North Hugosson, C.D. Berg, A. Auvinen, G.L. Andriole, M.J. Roobol,
Broadway, Baltimore, MD 21231. E.D. Crawford, V. Nelen, M. Kwiatkowski, M. Zappa, M. Lujan,
Dr. Auvinen: School of Health Sciences, PL 100 33014 Univer- A. Villers, E.J. Feuer, H.J. de Koning, A.B. Mariotto, R. Etzioni.
sity of Tampere, Finland. Provision of study materials or patients: J. Hugosson, G.L. An-
Dr. Andriole: Division of Urologic Surgery, Department of driole, E.D. Crawford, M. Lujan, A. Villers, H.J. de Koning.
Surgery, Washington University School of Medicine, 4921 Statistical expertise: A. Tsodikov, R. Gulati, P.F. Pinsky, S. Qiu,
Parkview Place, St. Louis, MO 63110. E.J. Feuer, R. Etzioni.
Dr. Roobol: Department of Urology, Erasmus Medical Center, Obtaining of funding: A. Tsodikov, G.L. Andriole, M. Lujan, A.
PO Box 2040, 3000 CA Rotterdam, the Netherlands. Villers, H.J. de Koning, A.B. Mariotto, R. Etzioni.
Dr. Crawford: Urologic Oncology, University of Colorado, Administrative, technical, or logistic support: A. Tsodikov, R.
1665 Aurora Court, Aurora, CO 80045. Gulati, C.D. Berg, E.J. Feuer, A.B. Mariotto.
Dr. Nelen: Provinciaal Instituut voor Hygiene, Kronenburg- Collection and assembly of data: A. Tsodikov, R. Gulati, P.F.
straat 45, 2000 Antwerpen, Belgium. Pinsky, S.M. Moss, S. Qiu, J. Hugosson, C.D. Berg, A. Auvinen,
Dr. Kwiatkowski: Department of Urology, Kantonsspital Aarau, G.L. Andriole, M.J. Roobol, E.D. Crawford, V. Nelen, M. Kwiat-
CH5001 Aarau, Switzerland. kowski, H.J. de Koning, A.B. Mariotto.
Dr. Zappa: Unit of Epidemiology, Institute for Cancer Preven-
tion, Via delle Oblate 2, 50141, Florence, Italy.
Dr. Lujan: Servicio de Urologa, Hospital Universitario Infanta
Cristina, Universidad Complutense de Madrid, Avenida
Nueve de Junio n 2, Parla, 28981 Madrid, Spain.
Dr. Villers: Department of Urology, CHU Lille, Universite de
Lille, F-59000 Lille, France.
Drs. Feuer and Mariotto: Division of Cancer Control and Pop-
ulation Sciences, National Cancer Institute, 9609 Medical Cen-
ter Drive, Bethesda, MD 20892.

Annals.org Annals of Internal Medicine

Downloaded From: https://annals.org/ by a Univ of California San Francisco User on 09/04/2017